Problem Statement

Learn what operator is doing and automate the process, making it a complete autonomous system.


  1. Maintaining/Controlling the Quality of the yield (output gas)
  2. Operating the plant to provide optimal performance from control loop level to total plant level. This includes learning the setpoints, highs, lows from the data and bottlenecks.
  3. Identify wear and tear of the equipment of the plant. Performance degradation needs to be identified to identify needs of maintenance.
  4. Reduce fuel consumption or increase the plant performance with same fuel costs.

Approach Methodology

  • Blend multiple datasets collected by PLC machines and plant operators, to form a unified "chromosome" representation. Identify data collection issues through cross-correlating information. Perform detailed descriptive analytics on the blended data up to very fine granularities, to identify Key Performance Statistics (KPS)
  • Use multiple machine learning models to predict key productivity concerns like maintenance events, downtimes and pipe rejections.

Modelling Approach

  • Descriptive analytics are used to drill down through the data at granularities that human experts typically cannot. We used models like ARX to mimic the existing plant and identify key insights on the existing plant operations
  • Once aberrations are identified using descriptive analysis, we developed predictive analytics to identify relationships between seemingly unrelated events.